Interactive Visualization and Analysis of Multidimensional Data Using Projections

Basic data for this project

Description

Nowadays, measured and simulated data are often multidimensional describing multiple characteristics of objects as scalar-, vector- or tensor-valued attributes simultaneously. Modern data analysis serves, e.g., to explore social processes, to find climatologic relations, or to test astrophysical hypotheses. When considering each data quantity separately, it is often not possible to get a deep insight into the interplay of dimensions and intrinsic data structure. Only a simultaneous analysis of all data components may help in discovering characteristic properties of the data. Consequently, visual representation and analysis of multidimensional data have gained increasing attention over recent years. An important part of modern visualization systems are various interactive tools allowing the user to bring his/her expertise into investigation of data at interactive frame rates. Projections are used for representing high-dimensional information in a low-dimensional domain (usually 2D), making visual analysis and interactions possible. Thus, developing of intuitive and understandable projections as well as interactive tools for efficient exploration, transformation and analysis of large data are in demand. Usability of interactive systems is characterized by the system response delays, which are usually negatively affected by large number of dimensions and large number of data samples. We propose methods for improving scalability of interactive systems by a proper grouping of control tools and visualized elements. We propose to involve the user in the process of clutter reduction, which shall reduce interaction times spent for accessing, selection and labeling samples in the projection layout. We develop intuitive and flexible algorithms for fast and efficient designing of customized projection layouts. Our approaches allow for direct multidimensional data manipulation in order to perform pre-projection data filtering and to improve projected data distribution. The proposed algorithms will allow for interactive and efficient discovery of significant relations and properties in complex datasets using visual analysis tools.